Hackers’ new goldmine: stolen AI access sold on the black market

Cybercriminals have found another way to exploit the growing demand for artificial intelligence: stealing companies' AI accounts and cloud credentials to run expensive AI workloads at someone else’s expense. The practice, known as LLMjacking, can generate more than $100,000 in cloud charges in just a few minutes, according to Google's Threat Intelligence Group (GTIG).

Hackers typically start by targeting publicly exposed AI endpoints or stealing API keys and cloud credentials, including identity tokens, service account keys, GitHub personal access tokens and active session tokens. Once inside a company's infrastructure, attackers attempt to escalate their privileges, establish persistent access and identify valuable AI resources.

Attackers ultimate goal is to monetize the stolen access by running their own AI workloads, routing requests through compromised accounts or reselling access to other users. Google describes the practice as the monetization and resource-hijacking of foundation models through token proxying, automated model streaming and GPU provisioning.

Unlike cryptojacking, which builds up costs gradually, LLMjacking can produce enormous financial damage almost immediately. Google says one group, tracked as UNC6993, generated more than $100,000 in billing charges after obtaining LLM API keys and enabling services inside a victim’s cloud project.

In another case, attackers subscribed compromised accounts to pay-as-you-go plans, selected premium AI models and launched high-volume inference workloads. To avoid rate limits, they repeatedly created and deleted short-lived service accounts while monitoring the victim’s billing dashboard.

Google says LLMjacking is developing into a specialized cybercrime ecosystem. Some groups specialize in breaking into corporate environments, while others — known as “LLM Mixers” — operate gateways that pool stolen credentials, route requests and resell unauthorized AI access to underground customers.

GTIG expects cybercriminals to increasingly prioritize LLMjacking as the market for AI services expands. Google urges organizations to improve visibility into their AI platforms and treat any unauthorized changes to AI accounts or services as a critical security event. It also recommends applying the principle of least privilege, ensuring users and accounts have only the permissions they need.